sohurdc opened a new issue, #9337:
URL: https://github.com/apache/paimon/issues/9337

   ### Search before asking
   
   - [x] I searched in the [issues](https://github.com/apache/paimon/issues) 
and found nothing similar.
   
   
   ### Paimon version
   
   1.3.1
   
   ### Compute Engine
   
   paimon-hive-connector-3.1, flink-1.17.1
   
   ### Minimal reproduce step
   
   ---
   labels: bug, hive, kerberos
   ---
   
   ## Describe the bug
   
   When a Paimon table is written through Flink and then queried from Hive via 
HiveServer2, the query fails with:
   
   ```
   SELECT * FROM paimon_sink LIMIT 1;
   Error: Error while compiling statement: FAILED: 
UnsupportedOperationException Proxy user is not supported 
(state=42000,code=40000)
   ```
   
   ## Environment
   
   - Paimon version: 1.3.1
   - Hive connector: `paimon-hive-connector-3.1`
   - Hive access mode: HiveServer2 (JDBC / Beeline) with proxy user 
(`hive.server2.proxy.user=bdwh`)
   - Kerberos: enabled
   
   ## Root cause analysis
   
   The error is thrown from `org.apache.paimon.security.KerberosLoginProvider`:
   
   ```java
   public void doLogin() throws IOException {
       if (principal != null) {
           LOG.info(
                   "Attempting to login to KDC using principal: {} keytab: {}", 
principal, keytab);
           UserGroupInformation.loginUserFromKeytab(principal, keytab);
           LOG.info("Successfully logged into KDC");
       } else if (!isProxyUser(UserGroupInformation.getCurrentUser())) {
           LOG.info("Attempting to load user's ticket cache");
           UserGroupInformation.loginUserFromSubject(null);
           LOG.info("Loaded user's ticket cache successfully");
       } else {
           throwProxyUserNotSupported();
       }
   }
   ```
   
   The Kerberos login logic currently has three branches:
   
   1. `principal` is non-null -> 
`UserGroupInformation.loginUserFromKeytab(principal, keytab)` (the Spark / 
Flink way).
   2. `principal` is null and the current user is **not** a proxy user -> 
`UserGroupInformation.loginUserFromSubject(null)` (load the ticket cache).
   3. `principal` is null and the current user **is** a proxy user -> throws 
`Proxy user is not supported`.
   
   The problem is that Hive cannot log into Kerberos the same way Spark does — 
Hive does **not** support passing `--keytab` directly. Instead, an 
already-Kerberos-authenticated HiveServer2 acts as a proxy on behalf of the 
business user, with the caller specifying `hive.server2.proxy.user=bdwh` (e.g. 
via Beeline).
   
   In this scenario, `principal` is null and the current UGI is a proxy user, 
so Paimon falls into branch 3 and throws `Proxy user is not supported`, which 
makes the whole query fail before it can even read the Paimon table.
   
   ## Proposed fix
   
   For the proxy-user case, Paimon should do nothing (the Kerberos credentials 
are already provided by the HiveServer2 delegation / proxy mechanism) instead 
of throwing.
   
   ```java
   public void doLogin() throws IOException {
       if (principal != null) {
           LOG.info(
                   "Attempting to login to KDC using principal: {} keytab: {}", 
principal, keytab);
           UserGroupInformation.loginUserFromKeytab(principal, keytab);
           LOG.info("Successfully logged into KDC");
       } else if (!isProxyUser(UserGroupInformation.getCurrentUser())) {
           LOG.info("Attempting to load user's ticket cache");
           UserGroupInformation.loginUserFromSubject(null);
           LOG.info("Loaded user's ticket cache successfully");
       } else {
           // Proxy user: do nothing, credentials are provided by HiveServer2 
proxy.
           // throwProxyUserNotSupported();
       }
   }
   ```
   
   Note: an earlier attempt replaced the throw with 
`UserGroupInformation.loginUserFromSubject(null)`, but that can cause TGT 
issues; the final fix is to simply do nothing for proxy users.
   
   ## Verification
   
   After applying the fix and rebuilding the Hive connector:
   
   ```bash
   mvn -DskipTests -Dspotless.check.skip=true clean package -pl paimon-hive -am
   # produces: paimon-hive-connector-3.1-1.3.1.jar
   ```
   
   The query succeeds:
   
   ```
   0: jdbc:hive2://dmc014011.venus.sohurdc.com:1> SELECT count(*) FROM 
paimon_sink LIMIT 1;
   +---------+
   | _c0     |
   +---------+
   | 154046  |
   +---------+
   1 row selected (75.586 seconds)
   ```
   
   Ranger authorization also works as expected (the proxy user still needs the 
required privileges):
   
   ```
   SELECT * FROM paimon_sink LIMIT 1;
   Error: Error while compiling statement: FAILED: HiveAccessControlException 
Permission denied: user [bdwh] does not have [SELECT] privilege on 
[bdwh/paimon_sink/*] (state=42000,code=40000)
   ```
   
   ## Expected behavior
   
   When Kerberos proxy user is used (as in HiveServer2), Paimon should not 
throw `Proxy user is not supported` and should allow queries to proceed.
   
   ## Additional context
   
   Related code reference: 
`paimon-common/src/main/java/org/apache/paimon/security/KerberosLoginProvider.java`
   
   
   ### What doesn't meet your expectations?
   
   FAILED: UnsupportedOperationException Proxy user is not supported
   
   ### Anything else?
   
   _No response_
   
   ### Are you willing to submit a PR?
   
   - [x] I'm willing to submit a PR!


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to